steady: read the heap once the device is idle and two reads agree

Under the suite's parallel load a frame read ~80 KB high: MoltenVK's completion handlers release a
finished command buffer on their own thread and lagged. Settled, 12 of 12 runs four at a time pass;
the frame's bound drops from 64 KB to 4 KB.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Orkun ÇAKILKAYA 2026-09-28 12:04:09 +03:00
parent 9b482d609a
commit 5ec13a073f

View file

@ -15,54 +15,69 @@ program Steady {
function scene_draw_casters(render3d_st: mut Render3dState, light_vp: floats) -> void { }
function stream_fill(s: Stream, cx: int, cz: int, band: int) -> void { }
# The heap once the device is idle and MoltenVK's completion handlers, which release a finished
# command buffer on their own thread, have caught up: two reads 20 ms apart that agree. Read at
# once, a machine busy with other work showed ~80 KB of buffers still being handed back.
function settled(render3d_st: Render3dState) -> long {
Vk.device_wait_idle(render3d_st.gvk_dev)
var last = Vk.heap_bytes()
for k in 0 .. 100 {
Time.sleep_us(20000)
let now = Vk.heap_bytes()
if now == last { return now }
last = now
}
return last
}
# bytes gained by the heap over n rounds of the buffer path, after a warm-up of its own
function buffer_rounds(render3d_st: mut Render3dState, b: int, n: int) -> long {
let before = Vk.heap_bytes()
let before = settled(render3d_st)
for i in 0 .. n {
gpu_buffer_upload(render3d_st, b, 4096 + (i % 16) * 4096, null, GPU_DYNAMIC)
gpu_buffer_free(render3d_st, b)
}
return Vk.heap_bytes() - before
return settled(render3d_st) - before
}
function frame_rounds(render3d_st: mut Render3dState, n: int) -> long {
let before = Vk.heap_bytes()
let before = settled(render3d_st)
for i in 0 .. n {
r3d_frame(render3d_st, float(i) / 60.0)
r3d_present(render3d_st)
}
return Vk.heap_bytes() - before
return settled(render3d_st) - before
}
# bytes gained over n loads and releases of one model: its parse, its meshes and its textures
function model_rounds(render3d_st: mut Render3dState, n: int) -> long {
let before = Vk.heap_bytes()
let before = settled(render3d_st)
for i in 0 .. n {
let m = gltf_load(render3d_st, "packages/ludic.lab/plate", "plate.gltf", "plate")
model_release(render3d_st, m)
}
return Vk.heap_bytes() - before
return settled(render3d_st) - before
}
# bytes gained over n actors placed and released (actor_release): the records come round again
function actor_rounds(render3d_st: mut Render3dState, m: Model, n: int) -> long {
let before = Vk.heap_bytes()
let before = settled(render3d_st)
for i in 0 .. n {
let a = actor_new(render3d_st, m)
actor_place(a, float(i % 7), 0.0, 0.0, 0.0)
actor_hide_part(a, "none", true)
actor_release(render3d_st, a)
}
return Vk.heap_bytes() - before
return settled(render3d_st) - before
}
# bytes gained over n parses of a glTF document, each freed whole (Json.free_all): strings too
function parse_rounds(text: string, n: int) -> long {
let before = Vk.heap_bytes()
function parse_rounds(render3d_st: Render3dState, text: string, n: int) -> long {
let before = settled(render3d_st)
for i in 0 .. n {
let doc = Json.parse(text)
Json.free_all(doc)
}
return Vk.heap_bytes() - before
return settled(render3d_st) - before
}
handler Boot(render3d_st: mut Render3dState) phase Start {
@ -87,13 +102,13 @@ program Steady {
actor_rounds(render3d_st, am, 20)
let grew_a = actor_rounds(render3d_st, am, 2000)
let text = Fs.read_text("packages/ludic.lab/plate/plate.gltf")
parse_rounds(text, 20)
let grew_p = parse_rounds(text, 200)
parse_rounds(render3d_st, text, 20)
let grew_p = parse_rounds(render3d_st, text, 200)
print(`steady: the buffer path gained {grew_b} bytes over 5000 rounds, the frame {grew_f} over 600, a glTF parsed and freed {grew_p} over 200, a model loaded and let go {grew_m} over 200, an actor placed and released {grew_a} over 2000`)
# a few KB of slack for what the system's own libraries keep (Metal's caches, the allocator)
var ok = grew_b < 16384
if not ok { print("steady: FAILED - releasing and making a buffer again leaves memory behind") }
if grew_f >= 65536 {
if grew_f >= 4096 {
ok = false
print("steady: FAILED - a frame leaves memory behind")
}